Professional Documents
Culture Documents
Dirichlet Series
Dirichlet Series
John E. McCarthy
Copyright
c 2018 John E. McCarthy
All rights reserved.
Contents
Preface v
Notation vii
Chapter 1. Introduction 1
1.1. Exercises 10
1.2. Notes 10
Chapter 2. The Prime Number Theorem 11
2.1. Statement of the Prime number theorem 11
2.2. Proof of the Prime number theorem 12
2.3. Historical Notes 21
Chapter 3. Convergence of Dirichlet Series 23
Chapter 4. Perron’s and Schnee’s formulae 29
4.1. Notes 35
Chapter 5. Abscissae of uniform and bounded convergence 37
5.1. Uniform Convergence 37
5.2. The Bohr correspondence 42
5.3. Bohnenblust-Hille Theorem 43
5.4. Notes 53
Chapter 6. Hilbert Spaces of Dirichlet Series 55
6.1. Beurling’s problem: The statement 55
6.2. Reciprocals of Dirichlet Series 59
6.3. Kronecker’s Theorem 60
6.4. Power series in infinitely many variables 61
6.5. Besicovitch’s Theorem 63
6.6. The spaces Hw2 66
6.7. Multiplier algebras of H2 and Hw2 69
6.8. Cyclic Vectors 74
6.9. Exercises 74
6.10. Notes 75
Chapter 7. Characters 77
iii
iv CONTENTS
v
Notation
N = {0, 1, 2, 3, . . . }
N+ = {1, 2, 3, 4, . . . }
Z = integers
Q = rationals
R = reals
C = complex numbers
P = {2, 3, 5, 7, . . . } = {p1 , p2 , p3 , p4 , . . . }
Pk = {p1 , p2 , . . . , pk }
Nk = {n ∈ N+ : all prime factors of n lie in Pk }
s = σ + it, s ∈ C, σ, t ∈ R
Ωρ = {s ∈ C; Re s > ρ}
π(x) = # of primes ≤ x
µ(n) = Möbius function
d(k) = number of divisors of k
dj (k) = number of ways to factor k into exactly j factors
φ(n) = PEuler totient function
Φ(s) = p∈P log ps
p
P
Θ(x) = p≤x log p
σc = abscissa of convergence
σa = abscissa of absolute convergence
σ1 = max(0, σc )
σu = abscissa of uniform convergence
σb = abscissa of bounded convergence
F (x) summatory function
Z T
− normalized integral
−T
εn = Rademacher sequence
E = Expectation
T = torus
tl tl
z r(n)P:= z1t1 . . . zP t1
l , where n = p1 · · · pl
B : P an z r(n) 7→ P an n−s
Q : an n−s 7→ an z r(n)
∞
T = infinite torus
vii
viii NOTATION
√
β(x) = P2 sin(πx) P
H2 = { ∞ n=1 an n
−s
: 2
n |an | < ∞}
Mult (X ) = {ϕ : ϕf ∈ X , ∀f ∈ X }
Mϕ : f 7→ ϕf
D∞ = infinite polydisk
E(ε, f ) ε-translation numbers of f
Hw2 = weighted space of Dirichlet series
Hw2 =weighted space of power series
P∞ −s
7→ n∈NK an n−s
P
QK : n=1 an n
ρ = Haar measure on T∞
`2 (G) = Hilbert space of square-summable functions on the group G
Xq = Dirichlet characters modulo q
L(s, χ) = Dirichlet L series
R θ+1 1
p
H∞ (Ω1/2 ) = {g ∈ Hol (Ω1/2 ) : [supθ∈R supσ>1/2 θ |g(σ + it)|p dt] p < ∞}
h RT i1/p
1 p
kf kHp = limT →∞ 2T −T
|f (it)| dt
4 The left-hand side is less than or equal to a constant times the right-hand side
≈ Each side is 4 the other side
ρA (x, y) = sup{|φ(y)k : φ(x) = 0, kφk ≤ 1}
H∞ = H ∞ (Ω0 ) ∩ D
E : f 7→ hf, gi i
gi = kλi /kkλi k
CHAPTER 1
Introduction
We shall prove in Theorem 2.19 that the zeta function has no zeroes
on the line {<s = 1}.
Notation 1.4. We shall let P denote the set of primes, and when
convenient we shall write
P = {p1 , p2 , p3 , p4 , . . . } = {2, 3, 5, 7, . . . }
to label the primes in increasing order. We shall let Pk denote the first
k primes.
Y −1 ∞
1 X
1− s = ζ(s) = n−s .
p∈P
p n=1
Formal proof:
1
P
Theorem 1.9. = ∞.
p∈P p
n
X n
X
ak b k = [Ak − Ak−1 ]bk
k=m k=m
Xn n
X
= Ak bk − Ak−1 bk
k=m k=m
n−1
X
= Ak [bk − bk+1 ] − Am−1 bm + An bn .
k=m
1. INTRODUCTION 5
Its values for the first few positive integer are in the table below:
n 1 2 3 4 5 6 7 8 9 10 11 12
µ(n) 1 -1 -1 0 -1 1 -1 0 0 1 -1 0
Theorem 1.14. For σ > 1,
∞
1 X
= µ(n)n−s .
ζ(s) n=1
1 Y 1
= 1− s
ζ(s) p∈P
p
1 1 1
= 1− s 1− s 1 − s ...
X2 3X 5
−s −s −s
= 1− p + p q − ...
p∈P p,q∈P, p6=q
∞
X µ(n)
=
n=1
ns
P∞
It is obvious that the Dirichlet series n=1 an n−s converges (con-
verges
P∞ absolutely, respectively) for all s ∈ Ωρ if and only if the series
−ρ −s
(a
n=1 n n )n converges (conv. abs., resp.) for all s ∈ Ω0 . This
ability to translate the Dirichlet series horizontally often allows one to
simplify calculations. (It is analogous to working with power series and
assuming the center is at 0). The proof of the proposition below is a
typical example of this.
The following “uniqueness-of-coefficients” theorem will be used fre-
quently.
Proposition 1.15. Suppose that ∞ −s
P
n=1 an n converges absolutely
to f (s) in some half-plane Ωρ and f (s) ≡ 0 in Ωρ . Then an = 0 for all
n ∈ N+ .
1. INTRODUCTION 7
where dj (k) denotes the number of ways to factor k into exactly j fac-
tors. Here, 1 is allowed to be a factor and two factorings that differ
only by the order of the factors are considered to be distinct.
Proof: We shall prove the first formula. Using Theorem 1.16, we
have, for σ > 1,
∞
! ∞ !
X X
ζ 2 (s) = n−s m−s
n=1 m=1
∞
X X
= 1 k −s
k=1 n|k
∞
X
= d(k)k −s .
k=1
Proof:
X X X
µ(q/p)g(d) = µ(q/p) f (n)
d|q d|q n|d
X X
= µ(q/d) f (n)
q q
d|q, d | n
n|q
X X
= µ(s) f (n)
q
n|q s| n
= f (q),
since, by the preceding corollary, the bracket is non-zero only when
q/n = 1.
Y p2
1 p
= 1− s 1 + s + 2s + . . .
p∈P
p p p
p2 p2
Y p 1 p
= 1 + s + 2s + . . . − s + 2s + 3s + . . .
p∈P
p p p p p
p2
Y 1 p
= 1+ 1− + + ...
p∈P
p ps p2s
X∞
= an n−s ,
n=1
where
k k
Y 1 r
Y 1
an = 1− qj j = n 1− = φ(n),
j=1
qj j=1
qj
for n = q1r1 . . . qkrk .
1.1. Exercises
1. Prove that if χ : N+ → T∪{0} is a quasi-character, which means
χ(mn) = χ(m)χ(n), the same argument that proved the Euler product
formula (Theorem 1.5) shows that
∞ Y
X χ(n) 1
= .
n=1
n−s p∈P
1 − χ(p)p−s
1.2. Notes
For a thorough treatment of the Riemann zeta function, see [Tit86].
The material in this chapter comes from the first few pages of Titch-
marsh’s magisterial book.
CHAPTER 2
and
π(109 ) = 50, 847, 478
π(109 )
109 =⇒ 109
≈ 1.05
≈ 48, 254, 942 log 109
log(109 )
Indeed, the LHS is a positive integer that is divisible by the RHS. Note
that we have not yet proved that there are any primes between n and
2n, so the RHS may be an empty product (we interpret empty products
as having the value 1).
Thus,
2n Y
≥ p = eΘ(2n)−Θ(n) .
n n<p≤2n
2.2. PROOF OF THE PRIME NUMBER THEOREM 13
Now
1 1 1
= s+ s s (2.9)
−1 p ps
p (p − 1)
Combining (2.8) and (2.9), we obtain, for s ∈ Ω1 ,
−ζ 0 (s) X log p X log p
= + (2.10)
ζ(s) p∈P
ps p∈P
ps (ps − 1)
Note that we can rearrange the terms since the series converge abso-
lutely in Ω1 . Thus, for s ∈ Ω1 ,
−ζ 0 (s) X log p
= Φ(s) + s (ps − 1)
(2.11)
ζ(s) p∈P
p
holds.
1 −1
Q
Proof: We have ζ(s) = p∈P (1 − ps
) and thus
ζ 0 (s) X p−s
= − log p
ζ(s) p∈P
1 − p1s
X log p
= −
p∈P
ps − 1
which proves that the first and last term in the statement of the propo-
sition are equal. For Re s > 1, k1/ps k < 1, so the first equality above
yields
ζ 0 (s) X 1 1
= − (log p)p−s (1 + s
+ 2s + . . . )
ζ(s) p∈P
p p
∞
XX
= − log p(pk )−s .
p∈P k=1
thus,
F 0 (x) X
Re = −Λ(n)n−x [3 + 4 cos(t0 log n) + cos(2t0 log n)]
F (x) n≥2
X
= −Λ(n)n−x [2 + 4 cos(t0 log n) + 2 cos(t0 log n)]
n≥2
Proof: Let
Z T
gT (s) = f (t)e−st dt . (2.23)
0
s2
Z
1 sT ds
g(0) − gT (0) = [g(s) − gt (s)] e 1+ 2 , (2.24)
2πi C R s
2
since est (1 + Rs 2 ) has value 1 at 0 and is
holomophic
everywhere in our
2
contour. Let h(s) := [g(s) − gt (s)] esT 1 + Rs 2 . In Ω0 , we have
∞
Z
−st
|g(s) − gT (s)| =
f (t)e dt
T
Z ∞
e−st
≤ M dt
ZT ∞
e−(Re
s)t
= M dt
T
1 −Re s T
= M e
Re s
Thus
Z Z
ds f (t)e−st dt
h(s) ≤
C+ s C+
e−Re sT
Z sT 2
e s
≤ M 1 + | ds |
C+ Re s s R2
s2 1 R 2 + s2 |s|2 + s
s+s
1+ 2 = 2
= 2
=
R s R s R s R2
18 2. THE PRIME NUMBER THEOREM
Thus,
e−Re sT eRe sT 2Re s
Z Z
ds M
h(s) ≤ | ds |
C+ s 2π C+ Re s s R2
M
≤ πR
πR2
M
=
R
We conclude C+ h(s) ds
R
s
→ 0 as R → 0.
For C− , we will show that both
s2
Z
sT ds
I1 (T ) := |g(s)| e 1+ 2
C− R s
and
s2
Z
sT ds
I2 (T ) := |gT (s)| e 1+ 2
C− R s
tend to 0 as R tends to 0.
We start with I1 :
T
Z
−st
|gT (s)| =
f (t)e dt
0
Z T
≤ M e−(Re s)t
dt
0
Z T
≤ M e−(Re s)t
dt
−∞
M −Re st
= e
Re s
Therefore,
M e−(Re s)t sT s2 1
Z
I1 (T ) ≤ e 1+ 2 | ds |
C− |Re s| R s
and since gT is an entire function, we can integrate over the semicircle
Γ− instead of C− and use the same estimates as in Ω0 to get
M
I1 (T ) ≤ .
R
Now
s2 1 sT
Z
I2 (T ) = g(s) 1 + 2 e ds
C− R s
2.2. PROOF OF THE PRIME NUMBER THEOREM 19
We are using the Stiltjes integral in the last expression because Θ(x)
is a step function.
We use integration by parts with u := x−s and dv := dΘ(x). Then
du = −sx−(s+1) dx and v(x) = Θ(x), giving
∞ Z ∞
−s Θ(x)
Φ(x) = x Θ(x) + s dx .
1 1 xs+1
The first term vanishes since Θ(x) = O(x) as x → ∞. We conclude
that Z ∞
Θ(x)
Φ(s) = s dx .
1 xs+1
Now let us use the substitution, x = et to get
Z ∞
Φ(s) = s Θ(et )e−ts dt .
0
We want apply Lemma 2.21 to f (t) := Θ(et )e−t −1 and g(s) = Θ(s+1)
s+1
−
1
s
. By Lemma 2.5, we get that f (t) is bounded and by Lemma 2.7, we
know that Θ(s+1)
s+1
− 1s is holomorphic in Ω0 . In order to apply Lemma
2.21, we need to check that g(s) is the Laplace transform of f (t).
We have
Z ∞ Z ∞
t −t −ts
Θ(e )e e dt = Θ(et )e−t(s+1) dt
0 0
and Z ∞
1
1e−ts dt =
0 s
20 2. THE PRIME NUMBER THEOREM
and
Z λxn
dt
= log(λxn ) − log xn = log λ
xn t
to conclude that cλ = Rλ − 1 − log λ > 0 by a well-known inequality for
∞
log. This implies that 1 Θ(x)−x
x2
dx does not converge, a contradiction.
The second case is that lim inf x→∞ Θ(x)x
< 1, so there is λ < 1 and
Θ(xn )
a sequence {xn } with xn → ∞ and xn < λ. As before,
xn λxn
Θ(t) − t λxn − t
Z Z
dt ≤ dt = −cλ = −(λ − 1 − log λ) < 0
λxn t2 xn t2
and we reach a contraction as in the first case. 2
2.3. HISTORICAL NOTES 21
= π(x) log x.
By Proposition 2.27, Θ(x) ∼ x and thus we have the bound
log x
lim sup π(x) ≥ 1.
x→∞ x
For the other bound, let ε > 0, and write
X
Θ(x) ≥ log p
x1−ε ≤p≤x
X
≥ log x1−ε
x1−ε ≤p≤x
During the proof of the Prime Number Theorem, we used the fact
that ζ(s) does not vanish for Re s ≥ 1. More precise estimates showing
that the zeros of ζ(s) must lie “close to” the critical line {Re s = 1/2}
yield estimates on the error |π(x) − Li(x)|.
The Riemann hypothesis is equivalent to the error estimate
√
π(x) = Li(x) + O( x log x).
The best known estimate is of the error is
A(log x)3/5
−
π(x) = Li(x) + O(xe (log log x)1/5 ).
CHAPTER 3
and M, N > n0
N
X N
X
−s
an n = (rn−1 − rn )n−s
n=M n=M
N −1
X 1 1
= rn s
− s (3.2)
n=M
(n + 1) n
rM −1 rN
+ s
− s.
M N
The absolute values of the last two terms are bounded by ε, since their
numerators are bounded by ε while the denominators have absolute
value at least 1. To estimate (3.2), note that
Z n+1
1 1 −s
s
− s = dx ,
(n + 1) n n xs+1
so that
Z n+1
1 1 dx |s| 1 1
(n + 1)s − ns ≤ |s| = − . (3.3)
n |xs+1 | σ nσ (n + 1)σ
Thus the absolute value of (3.2) satisfies, for M, N > n0 ,
N −1 N −1
X 1 1 X |s| 1 1
rn − ≤ |rn | −
(n + 1)s ns σ nσ (n + 1)σ
n=M n=M
N −1
|s| X 1 1
≤ ε −
σ n=M nσ (n + 1)σ
|s| 1 1
≤ ε −
σ Mσ Nσ
≤ c(δ)ε, (3.4)
since |s| = |1/ cos(arg s)| ≤ 1/ cos π2 − δ =: c(δ). This proves that
σ
the series is uniformly Cauchy, and hence uniformly convergent.
P∞
Corollary 3.5. If n=1 an n−s converges at s0 ∈ C, then it con-
verges in Ωσ0 .
S
Proof: This follows from the inclusion Ωσ0 ⊂ δ>0 {s : arg |s−s0 | <
π
2
− δ}.
This implies that there exists a unique value σc ∈ [−∞, ∞] such
that the Dirichlet series converges to the right of it, and diverges to the
left of it.
3. CONVERGENCE OF DIRICHLET SERIES 25
Proof: We will
Pshow (1); the second part has a similar proof. Hence
∞
we assume that n=1 an diverges and define
log |sn |
α := lim sup .
n→∞ log n
We will first prove the inequality α ≤ σc . Assume that ∞ −σ
P
n=1 an n
converges. Thus σ > 0Pand we need to show that σ ≥ α. Let
n
bn = an n−σ and Bn = k=1 bk (so that B0 = 0). By assumption,
the sequence {Bn } is bounded, say by M , and we can use summation
by parts as follows:
N
X
sN = an
n=1
N
X
= bn n σ
n=1
N
X −1
= Bn [nσ − (n + 1)σ ] + BN N σ
n=1
so that
N
X −1
|sn | ≤ M [(n + 1)σ − nσ ] + M N σ
n=1
σ
≤ 2M N .
Applying the natural logarithm to both sides yields
log |sn | ≤ σ log N + log 2M,
3. CONVERGENCE OF DIRICHLET SERIES 27
so
log |sn | log 2M
≤σ+ ,
log N log N
and this tends to σ as N → ∞, giving the desired upper bound for α.
We need to show thePother inequality: σc ≤ α. Suppose that σ > α;
we need to show that ∞ n=1 an n
−σ
converges. Choose an ε > 0 such
that α + ε < σ. By definition, there exist n0 ∈ N such that for all
n ≥ n0
log |sn |
≤ α + ε.
log n
This implies that
log |sn | ≤ (α + ε) log n = log(nα+ε ).
Thus, |sn | ≤ nα+ε , for all n ≥ n0 . Observe that
Z n+1
1 1 du
σ
− σ
= σ σ+1
≤ σn−(σ+1) .
n (n + 1) n u
Using summation by parts, we can compute
N N
X an X
sn n−σ − (n + 1)−σ + sN (N + 1)−σ − sM M −σ
=
n=M +1
nσ n=M
XN
nα+ε σn−σ−1 + N α+ε N −σ + M α+ε M −σ
≤
n=M
. (M − 1)α+ε−σ ,
and the last quantity tends to zero as M tends toR∞.
N −1
We estimated N α+ε−σ−1
by the integral M −1 xα+ε−σ−1 dx .
P
n=M n
(M −1)α+ε−σ , and the symbol . means less than or equal to a constant
times the right hand-side (where the constant depends on α + ε − σ,
but, critically, not on M ).
Suppose you know the function values f (s) of some function f that,
at
P∞ least in some half-plane, can be represented by the Dirichlet series
−s
n=1 an n . How do you determine the coefficients an ? We have seen
one way already in Proposition 1.15:
a1 = lim f (s)
s→∞
a2 = lim 2s [f (s) − a1 ]
s→∞
a3 = lim 3s [f (s) − a1 − a2 2−s ]
s→∞
for all λ ∈ R.
Proof: See [Tit37, Thm. 24].
This gives us the following formula for the inverse of the Mellin
transform.
Theorem 4.4. Suppose that g ∈ BVloc (0, ∞). Let σ ∈ R, and
assume that g(x)xσ−1 ∈ L1 (0, ∞). Then
Z σ+iT
1 + − 1
(Mg)(s)x−s ds,
g(x ) + g(x ) = lim
2 2πi T →∞ σ−iT
for all x > 0.
Proof: Let λ = log x, then G(λ) := g(eλ ) belongs to BVloc (−∞, ∞).
Let h(λ) := G(λ)eλσ . Then
Z ∞ Z ∞
|h(λ)| dλ = |g(eλ )|eλ(σ−1) eλ dλ
−∞
Z−∞∞
= |g(x)|xσ−1 dx
0
< ∞,
so h belongs to L1 (−∞, ∞). It also belong to BVloc (−∞, ∞), be-
cause, locally, it is the product of a function of bounded variation and
a bounded increasing function. We have
Z ∞
(Mg)(s) = g(x)xs−1 dx
Z 0∞
= G(λ)eλs dλ
Z0 ∞
G(λ)eλσ eiλt dt
=
−∞
= F G(λ)eλσ (−t)
= (Fh)(−t).
Now, we apply Theorem 4.3 to h.
1 + 1
g(x ) + g(x− ) = e−λσ h(λ+ ) + h(λ− )
2 2 Z T
−λσ 1
= e lim (Fh)(ξ)eiλξ dξ
2π T →∞
Z−T
T
1
= e−λσ lim (Mg)(σ − it)eiλt dt
2π T →∞ −T
Z T
−λσ 1
= e lim (Mg)(σ + it)e−iλt dt
2π T →∞ −T
4. PERRON’S AND SCHNEE’S FORMULAE 31
Z T
1
= lim (Mg)(σ + it)e−λ(σ+it) dt
2π T →∞ −T
Z σ+iT
1
= lim (Mg)(s)e−λs ds
2πi T →∞ σ−iT
Z σ+iT
1
= lim (Mg)(s)x−s ds,
2πi T →∞ σ−iT
and we are done.
P∞ P0
Given a Dirichlet series n=1 an n−s , let F (x) = n≤x an , where
P0
means that for x = m ∈ N+ , the last term of the sum is replaced
am
by 2 so that the function F (x) satisfies
1
F (x+ ) + F (x− )
F (x) =
2
for all x. This function F (x) is called the summatory function of the
Dirichlet series.
Exercise 4.16. Check that the same proof yields Schnee’s theorem
for generalized Dirichlet series. Let λn be a strictly increasing sequence
with
P lim n→∞ λn = ∞. Define the abscissa of convergence for f (s) =
an e−λn s just as for an ordinary Dirichlet series. Then, for σ > σc ,
Z T (
1 iµt an e−λn σ , if µ = λn ,
lim f (σ + it)e dt = (4.17)
T →∞ 2T −T 0, otherwise.
4.1. Notes
Perron’s formula, like Schnee’s theorem, also holds for generalized
Dirichlet series. For further results in this vein, see [Hel05, Ch. 1].
CHAPTER 5
We will show that the contribution of the tail of the series above
— the sum for n > N — is small, while the sum over n ≤ N is
approximately the partial sum of the Dirichlet series.
First, assume that n > N , i.e., n ≥ N + 1. Apply Cauchy’s theo-
rem to the rectangular path whose left-hand edge is RHE and whose
horizontal sides have length L, and let L tend to infinity. Since the in-
tegrand has no poles in the region encompassed by this rectangle, the
integral over the closed path vanishes. On the new right-hand edge,
the integrand decays exponentially with L and so the limit of the inte-
gral over this edge tends to 0. On the top edge (and similarly, on the
bottom one), we estimate as follows,
Z
∞+it±iN d N + 1 z−s dz 1
Z ∞
N + 12
x−σ
2
dz ≤ dx
n z−s N d σ+(σa −ρ) n
s+(σa −ρ)±iN d
N+ 1
Z ∞
1 (x−σ) log n
2
= e dx
N d σ+(σa −ρ)
N+ 1
1 1 (σa −ρ) log n
2
= e .
N d − log N + 12
n
N + 12
The expression log n
is minimized when n = N + 1. So
1
N+ 2 1
| log | ≥ − log(1 − )
n 2N + 2
1
> .
2(N + 1)
5.1. UNIFORM CONVERGENCE 41
. N −1 ,
P |an |
since nσ+σa −ρ
converges.
If n ≤ N , we use Cauchy’s theorem again, but now with a rectan-
gular path whose right-hand edge is RHE and whose width L tends to
infinity. The residue theorem now implies that
z−s
N + 12
Z
1
dz = 2πi.
n z−s
The integrand decays exponentially on the left-hand edge, and so the
integral over that edge tends to zero. As for the top edge (and also the
bottom one)
Z
s+(σa −ρ)±iN d N + 1 z−s dz 1
Z σ+(σa −ρ)
N + 12
x−σ
2
dz ≤ dx
n z−s N d −∞ n
−∞+it±iN d
N+ 1
Z σ+(σa −ρ)
1 (x−σ) log n
2
= e dx
N d −∞
N+ 1
1 1 (σa −ρ) log n
2
= e
N d log N + 12
n
σ −ρ
N + 12 a
−d 1
≤ N 1
N+
log N 2 n
σ −ρ
N + 12 a
1−d
. N
n
. N −1 n−σa +ρ
Thus,
N 1 z−s N N
N+ |an |
Z
X
−s 2 1 X an −1 −σa +ρ
X
an n dz = 2πi + O(N n )
n=1 RHE n z−s n=1
ns n=1
nσ
N
X an
= 2πi + O(N −1 ),
n=1
ns
42 5. ABSCISSAE OF UNIFORM AND BOUNDED CONVERGENCE
For finite series, we can norm both spaces so that B will be isomet-
ric. Indeed, we have
N N
Z T
X 2
1 −it
X
lim an n dt = |an |2
T →∞ 2T −T
n=1 n=1
Z N 2
X 2πit·r(n)
=
a n e dt. (5.9)
T ∞
n=1
∞
By T we mean the infinite torus
T∞ = {(e2πit1 , e2πit2 , . . . ) : 0 ≤ tj < 1 ∀j ∈ N+ }
which we identify with the infinite product
[0, 1) × [0, 1) × · · ·
on which we put the product probability measure of Lebesgue measure
on each interval.
We shall investigate when (5.9) holds for infinite sums in Theorem
6.39.
Flesh this section out.
x2
arbitrary real number. Then, using the inequality 12 (ex + e−x ) ≤ e 2
yields
P
E eλQ(t) = E eλ n εn Pn (t)
!
Y
= E eλεn Pn (t)
n
Y
E eλεn Pn (t)
=
n
Y 1
λPn (t) −λPn (t)
= [e +e ]
n
2
2
2 Pn (t)
Y
≤ eλ 2
n
λ2
kPn k2∞
Y
≤ e 2
n
λ2
kPn k2∞
P
= e 2 n
τ λ2
= e 2 . (5.18)
By Lemma 5.14, there exists an interval I = I(ε) ⊂ Tr of volume at
least (N + 1)−2r such that |Q| ≥ 12 kQk∞ of I. For fixed ε = {εn } we
thus have
Z
λM (ε) 1
e 2 ≤ eλQ(t) + e−λQ(t) dt
vol(I(ε)) ZI(ε)
≤ (N + 1)2r eλQ(t) + e−λQ(t) dt
Tr
τ λ2
+log 2+2r log(N +1)
= e 2
Thus,
λM λ2 τ
E e 2 − 2 −log 2−2r log(N +1) ≤ 1,
5.3. BOHNENBLUST-HILLE THEOREM 49
Proof: For any γ ≥ 8, the probability that a random series will not
have the property is at most γ2 < 1.
Consequently,
N 2
2
X −(ρ+it)
K ≥
a n n
n=1
N
m
X X
= |an |2 n−2ρ + 2 Re an am (nm)−ρ eit log n .
n=1 1≤n<m≤N
5.3. BOHNENBLUST-HILLE THEOREM 51
so that
X 12
k 2
kQk k∞ ≤ C 2 log k |εj | .
|j|=k
Claim 1: σa (f ) = 1.
52 5. ABSCISSAE OF UNIFORM AND BOUNDED CONVERGENCE
≤ p−kσ
2k
sup |Qk |
|zi |=1
1
2k + k − 1 2
≤ Cp−kσ
2k
2 log kk
(5.28)
k
h i1
k −kσ k k2 2
. (2 k log 2) 2 log k2
2 1 1 p
= (k log 2)−kσ 2k (−σ+ 2 + 2k ) log k
2 1 p
= (k log 2)−kσ 2k (−ε+ 2k ) log k.
P
The
P series k |fk | is thus estimated by a summable series. Hence,
k fk converges to a holomorphic function which is bounded in Ω1/2+ε
5.4. NOTES 53
Lemma
n+m−1
5.29. The number of monomials of degree m in n variables
is m
.
Proof: From a linear array of n + m − 1 objects, choose n − 1 and
color them black. Let the power of zi be the number of non-colored
objects between the (i − 1)st black one and the ith one. 2
Exercise 5.30. Fill in the details that the series in (5.28) con-
verges.
Exercise 5.31. Show that for all x ∈ [0, 12 ], there is a Dirichlet
series such that σa − σu is exactly x.
(Hint: Although Bohnenblust and Hille did not spot it, this result
is a one-line consequence of Theorem 5.26. If you find the right line!)
5.4. Notes
The proofs of the Bohnenblust-Hille theorem in Section 5.3 and
Bohr’s Theorem 5.4 are based on H. Boas’s article [Boa97]. The orig-
inal proofs are in [BH31] and [Boh13b], respectively. Theorem 5.24
was proved in [Boh13a].
Talk about recent advances, in particular [DFOC+ 11].
CHAPTER 6
{β(nx) : n ∈ N+ }
and
2
X
X
a e
n n
= |an |2 .
n n
Thus condition (6.3) is equivalent to boundedness of T from below and
above. Moreover, T is onto if and only if the span of {vn } is dense in
H. The claim follows by recalling that a map is invertible if and only
if it is bounded, bounded from below, and onto.
Here is Beurling’s question.
Question 6.5. (Beurling) For which odd 2-periodic functions ψ :
R → C does the sequence {ψ(nx)}∞ 2
n=1 form a Riesz basis for L ([0; 1])?
whenever 2σ > 1. In fact, the above estimate shows that for any
s0 ∈ Ω1/2 , the map H2 3 f 7→ f (s0 ) is a bounded linear functional.
Therefore it is given by the inner product with a function ks0 ∈ H2 ,
the so-called reproducing kernel at s0 , i.e.,
f (s0 ) = hf, ks0 i for all f ∈ H2 .
For any Hilbert (or Banach) space of analytic functions X , we define
its multiplier algebra by
Mult (X ) = ϕ; ϕf ∈ X , ∀f ∈ X .
It is easy to check that the following hold
• 1 ∈ X =⇒ Mult (X ) ⊂ X ,
• Mult (X ) is an algebra.
Clearly, multiplication by k −s is isometric on H2 , for all k ∈ N. Con-
sequently, every finite Dirichlet series lies in Mult (H2 ).
1
Also, note that sups∈Ω1/2 |k −s | = k − 2 → 0 as k tends to ∞. Thus,
kf kMult (H2 ) 6. kf kH ∞ (Ω1/2 ) .
The following result — multiplication operators are bounded if they
are everywhere defined — is true in great generality (see Section 11.4).
Proposition 6.9. The multiplication operator Mϕ is bounded on
H2 for every ϕ ∈ Mult (H2 ).
Proof: Multiplication operators on a Banach space of functions in
which norm convergence implies pointwise convergence (or at least a.e.
convergence) are easily seen to be closed. Indeed, suppose that fn → f
and Mϕ fn → g. Then, for every s ∈ Ω1/2 , fn (s) → f (s) and so
(Mϕ fn )(s) = ϕ(s)fn (s) → ϕ(s)f (s) = (Mϕ f )(s). On the other hand,
Mϕ fn (s) → g(s), for all s ∈ Ω1/2 . We conclude that (Mϕ f )(s) = g(s)
for all s ∈ Ω1/2 and hence Mϕ f = g. Thus, Mϕ is closed. Hence,
Mϕ is an everywhere defined closed linear operator on a Banach space,
and the closed graph theorem states that such operators are necessarily
bounded.
58 6. HILBERT SPACES OF DIRICHLET SERIES
We have
X X X
−s −s
kgf k2H2 = cn ak (nk) , cm aj (mj) = ak c n aj c m ,
n,k m,j k,n,j,m; kn=jm
and so the condition (6.9) holds, if and only if kgf kH2 ≈ kf kH2 , i.e.,
when Mg is bounded and bounded below.
Let us also look the density of the span of {ψn }n . It is equivalent
to
X X
2
span cn β(nkx) = L ([0; 1]) ⇐⇒ span cn enk = `2 (N)
n k∈N+ n k∈N+
X
−s
⇐⇒ span cn (nk) = H2
k∈N+
n
⇐⇒ span k −s g(s) = H2 .
k∈N+
m
j’s yield distinct γj ’s. Also, |j|=m |aj | = (k + 1) , since there are
(k + 1) terms, each with a coefficient of modulus 1.
Suppose that lim supt→∞ F (t) < k + 1. Then there exist M > 0
and λ < k + 1 such that |F (t)| ≤ λ for all t > M . Consequently,
1 T
Z
lim sup |F (t)|m dt ≤ λm .
T →∞ T 0
1 T
Z
≤ lim sup |F (t)|m dt
T →∞ T 0
≤ λm . (6.15)
Note that there are m+k
k
≤ (m + 1)k possible j’s. Thus, summing the
inequality (6.15) over all j’s yields
X
(k + 1)m = |aj |
|j|=m
≤ (m + 1)k λm ,
a contradiction for large m.
to D. So ψz is a quasi-character.
Hilbert, in 1909, asked:
Question 6.19. Does Qf (z) make sense on a larger set than D∞ ∩
2
`?
This was his answer. Let z = (z1 , z2 , . . . ), and let z(m) denote
(z1 , . . . , zm , 0, 0, . . . ). Consider the sequence Fm (z) := F (z(m) ); this is
called the mte -Abschnitt (or cut-off). If f ∈ H2 and F = Qf , then the
functions Fm are well-defined on D∞ by Proposition 6.17.
Proposition 6.20. (Hilbert) Suppose that there exists C > 0
such that
|Fm (z)| ≤ C ∀ z ∈ D∞ , ∀ m ∈ N+ .
Then, for every z ∈ D∞ ∩ c0 , the limit
lim Fm (z) =: F (z)
m→∞
exists.
6.5. BESICOVITCH’S THEOREM 63
θ2 = log
2π
3
and α1 = α2 = 0. Then there exists an arbitrarily large τ ∈ R
so that
τ log 2 τ log 3
dist ,Z < ε and dist , Z < ε.
2π 2π
Thus,
|2−(s+iτ ) − 2−s | = |2−s (e−iτ log 2 − 1)|
−ρ τ log 2
≤ 2 2π(log 2) dist ,Z
2π
< Cε.
Similarly, one obtains
τ log 3
|3−(s+iτ ) − 3−s | ≤ 3−ρ 2π(log 3) dist { , Z} < Cε.
2π
There exists a refined version of Kronecker’s theorem that implies that
the ε-translation numbers of f are relatively dense, so f is uniformly
almost periodic. However, the claim also follows from Corollary 6.28
below.
Theorem 6.25. (Besicovitch) Suppose f (s) = ∞ −s
P
n=1 an n and
the series converges uniformly in Ωρ . Then f is uniformly almost pe-
riodic.
Lemma 6.26. Suppose f is uniformly almost periodic and uniformly
continuous in Ωρ , and let 0 < ε1 < ε2 be arbitrary. Then there exists
a δ > 0 such that for each τ ∈ E(ε1 , f ), the inclusion (τ − δ, τ + δ) ⊂
E(ε2 , f ) holds.
Proof: Let δ > 0 be such that for every 0 < δ 0 < δ and z ∈ Ωρ ,
|f (z + iδ 0 ) − f (z)| < ε2 − ε1 .
For any τ 0 ∈ (τ − δ, τ + δ), write τ 0 = τ + δ 0 with 0 < |δ 0 | < δ. Then
the inequality
|f (z + iτ 0 ) − f (z)| ≤ |f (z + i(τ + δ 0 ) − f (z + iτ )| + |f (z + iτ ) − f (z)|
< (ε2 − ε1 ) + ε1 = ε2
holds.
Remark 6.35. On the unit disk, one can define spaces Hw2 by
X X
2 n 2
Hw := an z : |an | wn < ∞ . (6.36)
n n
A special case is when
wn = (n + 1)α .
Then α = 0 corresponds to the Hardy space, α = −1 to the Bergman
space, and α = 1 to the Dirichlet space, the space of functions whose
derivatives are in the Bergman space. The theory of the Hardy space
on the disk is fairly well-developed – see e.g. [Koo80, Dur70] for a
first course, or [Nik85] for a second. The Bergman space (and the
other spaces with α < 0 in this scale, that all come from L2 -norms
of radial measures) is more complicated — see e.g. [DS04, HKZ00].
The Dirichlet space on the disk is even more complicated analytically,
though it does have the complete Pick property. See e.g. [EFKMR14].
This section should be seen as an attempt to continue the analogy
of Remark 6.35. The case α = 0 in (6.34) we think of as a Hardy-type
space, and the case α = −1 in (6.34) we think of as a Bergman-type
space. When α > 0, we can still define weights by (6.34), though they
do not come from a measure as in (6.33). By Remark 6.30, we can think
of α = 1, for example, as the space of functions whose first derivatives
lie in the space with α = −1. This would render this space a “Dirichlet
space” of Dirichlet series, which is perhaps a surfeit of Dirichlet.
If the weights are defined by (6.33), then, for every ε > 0,
Z ε
wn ≥ n−2σ dµ
0
≥ µ([0, ε])n−2ε , (6.37)
and, consequently, the weight sequence cannot decrease to 0 very fast.
Proposition 6.38. Suppose wn is a weight sequence that is bounded
below by n−2ε for every ε > 0. Then for any f ∈ Hw2 , we have σa (f ) ≤
1
2
.
Proof: Take σ > 12 , and choose ε > 0 such that σ − ε > 12 . Then,
by the Cauchy-Schwarz inequality,
X X
|an |n−σ = |an n−ε |n−(σ−ε)
n n
! 21
X X 21
≤ |an |2 n−2ε n−2(σ−ε) .
n
68 6. HILBERT SPACES OF DIRICHLET SERIES
The first term is finite by (6.37), and the second since 2(σ − ε) > 1.
The following theorem, in the case that µ = δ0 , is due to F. Carlson
[Car22]. If w1 < ∞, we assume that the Dirichlet series for f starts
at n = 1; if w1 is infinite, we start the series at n = 2 (Condition (6.32)
says that w2 < ∞).
Theorem 6.39. Let µPsatisfy (6.31) and (6.32), and define wn by
(6.33). Assume that f = n an n−s has σb (f ) ≤ 0. Then
Z T Z ∞
X
2 1
|an | wn = lim lim |f (s + c)|2 dµ(σ) dt. (6.40)
c→0+ T →∞ 2T −T 0
n
Then
Z T Z 1/c Z T Z 1/c
1 1 X
lim 2
|f (s + c)| dµ(σ) dt = lim | an n−s−c |2 dµ(σ) dt + O(ε)
T →∞ 2T −T 0 T →∞ 2T −T 0 n≤N 0
X Z 1/c
= 2
|an | n−2σ−2c dµ(σ) + O(ε)
n≤N 0 0
then {wn }n is non-increasing and, since the measure is radial, the se-
quence {z n }n∈N is an orthogonal basis of Hw2 . (Saying the measure is
radial means dµ = dθdν(r) for some measure ν on [0, 1]). Thus, the
norm on Hw2 is given by integration:
Z
2
kf k = |f (z)|2 dµ(z).
D
the circle, the theorem is still true, and can most easily be seen by
writing Z Z
|f (z)| dµ(z) = lim |f (rz)|2 dµ(z).
2
D r%1 D
In particular, (6.44) holds for all the spaces with wn = (n + 1)α for
α ≤ 0.
Remark 6.45. There exist many functions in H ∞ (Ω0 ) \ D, for
−s
example f (s) = 23 s
and g(s) = (s+1)2.
Remark 6.47. The lemma does not hold without the assumption
that f is bounded in the strip. Indeed, consider the function f (z) =
iz
ee . It is holomorphic in the strip {− π2 < Re z < π2 }, bounded on
its boundary {|Re z| = π2 }, but limt→−∞ f (it) = ∞. However, one
can weaken the assumption of boundedness of f to an appropriate
restriction on the growth of f .
The following lemma is trivial if 1 ∈ Hw2 .
Lemma 6.48. Any multiplier of Hw2 lies in D.
Proof: If ϕ belongs to Mult (Hw2 ), then both ϕ(s)2−s and ϕ(s)3−s
are in D. So
X
ϕ(s)2−s = an n−s
X
ϕ(s)3−s = bn n−s .
Multiplying the first equation by 3−s and the second by 2−s , we con-
clude that an is zero when n is odd (and bn is zero when n is not
divisible by 3), so ϕ itself can be represented by an ordinary Dirichlet
series.
P∞ −s
Proposition 6.49. Let ϕ(s) = n=1 bn n with σb ≤ 0. Then
kMϕ k = kϕkΩ0 .
6.7. MULTIPLIER ALGEBRAS OF H2 AND Hw
2
71
Z T Z ∞
kϕf k2Hw2 = lim lim − |ϕ(s + c)|2 |f (s + c)|2 dµ(σ) dt
c→0+ T →∞ −T 0
≤ kϕk2Ω0 · kf k2Hw2 .
Hence Mϕ is bounded on a dense subset of Hw2 , and therefore extends
to a bounded operator on all of Hw2 , which must be multiplication by
φ. (Why?) Also, the estimate above shows that kMϕ k ≤ kϕkΩ0 .
Conversely, assume that kMϕ k = 1 and 1 < kϕkΩ0 (possibly infi-
nite). Let
N (σ) := sup |ϕ(σ + it)|.
t∈R
For K ∈ N+ , define
NK = {n = pr11 · . . . · prKK ; rj ∈ N},
where,
as usual, pl is the l-th prime. Clearly, n ∈ NK , if and only if
pl 6 n for all l > K. Let QK : D → D be the map defined by
X ∞ X
−s
QK an n = an n−s .
n=1 n∈NK
P∞
The map QK is well-defined, since if n=1 an n−s converges absolutely
in Ωρ , then so does n∈NK an n−s .
P
We need the following observations.
Lemma 6.52. For any K ∈ N+ , the map QK has the following
properties:
(1) The restriction of QK to Hw2 is the orthogonal projection onto
span {n−s : n ∈ NK }.
(2) For any ϕ, f ∈ D, QK (ϕf ) = (QK ϕ)(QK f ).
(3) If ϕ ∈ Mult (Hw2 ), then QK Mϕ QK = MQK ϕ QK = QK Mϕ .
Proof: (1) follows immediately from the orthogonality of the func-
tions {n−s }n∈N+ .
(2) By linearity, we only need to check that QK (n−s m−s ) =
QK (n−s )QK (m−s ), for +
all m, n ∈ N . This follows from the facts that
if p is prime, then p6 nm if and only if p6 n and p6 m, and
(
−s n−s , pl 6 n, for all l > K,
QK n =
0, otherwise.
6.7. MULTIPLIER ALGEBRAS OF H2 AND Hw
2
73
6.9. Exercises
Exercise 6.57. Show that H2 contains a function f with σa (f ) =
1
2
.
Exercise 6.58. Prove that the reproducing kernel for Hw2 is given
by
X 1
k(s, u) = n−s−ū . (6.59)
n
w n
Characters
Proposition
P∞ 7.11. Let f ∈ H2 , and let χ be a character. Then
fχ (s) := n=1 an χ(n)n−s is a vertical limit function of f . Conversely,
all vertical limit functions have this form for some character χ.
Proof: Fix a character χ and let k ∈ N+ . By Kronecker’s theorem,
we can find τk ∈ R such that |eiτk log pj − χ(pj )| ≤ 1/k holds for j =
1, . . . , k. Define fk := fiτk . Then using inequality (7.5), we conclude
r (n) r (n)
that for any n ∈ N+ , n = p11 . . . pl l ,
fbk (n) − fbχ (n) = fˆ(n)niτk − fˆ(n)χ(n)
80 7. CHARACTERS
l l
Y i log pj τ rj Y
fˆ(n)·
rj
= e k
− χ(p j )
j=1 j=1
l
X
≤ kf kH2 rj ei log pj τk − χ(pj )
j=1
l
1 X
≤ kf kH2 rj ,
k j=1
holds.
dk (n) = O(nε ) as n → ∞.
7.1. VERTICAL LIMITS 81
Z T ∞
1 X
lim k 2
|ζ̃ (σ + it)| dt = n−2σ |ζb̃k (n)|2
T →∞ 2T −T n=1
X∞
≤ n−2σ |ζbk (n)|2
n=1
< ∞,
since ζbk (n) = dk (n) = O(nε ) for all ε > 0 by Lemma 7.13. Thus we
would have proved the Lindelöf hypothesis. Conversely, the following
is known.
Theorem 7.14. (Titchmarsh) If
Z T
1
lim |ζ̃ k (σ + it)|2 dt < ∞,
T →∞ 2T −T
P∞
then it equals to n=1 n−2σ |ζb̃k (n)|2 .
82 7. CHARACTERS
For any function f ∈ H p (Ω0 ) and almost every t ∈ R, the limit f˜(it) :=
limσ→0+ f (σ + it) exists and satisfies f˜ ∈ Lp (iR). One can recover f
from f˜ by convolution with the Poisson kernel. For both Hip (Ω0 ) and
H p (Ω0 ) we identify the functions with their boundary values.
7.2. HELSON’S THEOREM 83
f (z)
f ∈ Hi2 (Ω0 ) ⇐⇒ ∈ H 2 (Ω0 ).
1+z
1−it k
R
Thus, the function G(χ) := R fχ (it)( 1+it ) dµ(t) belongs to L2 (K) ⊂
L1 (K). To show G(χ) = 0 for a.e. χ, we only need to show that all its
Fourier coefficients vanish, i.e,
Z
G(χ)χ(q) dρ(χ) = 0, for all q ∈ Q+ .
K
k
Remark 7.17. Recall that by Lemma 7.13, ζ1/2+ε and, conse-
k
quently, ζ̃1/2+ε belong to H2 for every k ∈ N+ and ε > 0. Thus,
by Theorem 7.15, for a.e. χ ∈ K
Z T 2k
1
lim − ζχ
+ ε + it dt < ∞,
T →∞ −T 2
and
Z T 2k
ζ̃χ 1 + ε + it dt < ∞.
lim −
T →∞ −T
2
The dual group of G, denoted by Ĝ, is the set of characters, i.e. the
multiplicative functions from G to T.
Proposition 7.22. Let G be a finite abelian group. Then Ĝ forms
an orthonormal basis of `2 (G).
Fix q, and let G := Z∗q . Any character e ∈ Ĝ = Z
c∗ extends to Z by
q
(
e(n mod q), if gcd(n, q) = 1,
e(n) =
0, otherwise.
Then e : Z → T ∪ {0} is totally multiplicative. Any such function is
called a Dirichlet character modulo q. We denote the set of all Dirichlet
characters modulo q by Xq . The trivial Dirichlet character modulo q
is the periodic extension of the trivial character on Z∗q , that is,
(
1, if gcd(n, q) = 1,
χ0 (n) =
0, otherwise.
We will identify Dirichlet characters modulo q with their restriction to
Z∗q .
88 7. CHARACTERS
1 X X χ(p)
= χ(l)
φ(q) χ∈X p∈P
ps
q
" #
1 X χ0 (p) X X χ(p)
= + χ(l) (7.17)
φ(q) p∈P ps χ6=χ p∈P
p s
0
Note that
X
log L(s, χ) = − log(1 − χ(p)p−s )
p∈P
X χ(p)
−2s
= − − −s + O(p )
p∈P
p
X χ(p)
= + O(1).
p∈P
p−s
Therefore, to prove Theorem 7.21, it is enough to show that
lims→1+ L(s, χ) is finite and non-zero, for every non-trivial Dirichlet
character χ. If q = q1r1 . . . qkrk , then
Y 1
L(s, χ0 ) = −s
= (1 − q1−s ) . . . (1 − qk−s )ζ(s).
p∈P
1 − χ0 (p)p
∞
!
XXX χ(pk )
= exp
χ
p∈P k=1
kpks
∞
!
XX 1 X
= exp χ(pk ) , (7.18)
p∈P k=1
kpks χ
1
= .
φ(q)
Consequently, 1 = φ(q)hδ1 , χi, so that for any n ∈ N+ , we obtain
X X
χ(n) = φ(q) hδ1 , χiχ(n)
χ χ
= φ(q)δ1 (n)
≥ 0.
Q
We conclude by (7.18) that χ L(s, χ) = exp(r), where r is non-
negative.
Suppose that L(1, η) = 0. Then L(s, η) = O(s − 1) as s tends
to 1. Its conjugate η is also a character (and different from η, since
we assumed η takes on some non-real value somewhere). Moreover,
L(1, η) = ∞ η(n)
P
n=1 n = L(1, η) = 0. Thus, as s → 1+,
Y Y
L(s, χ) = L(s, χ0 ) · L(s, η) · L(s, η) · L(s, χ)
χ χ6=η,η,χ0
= O((s − 1)−1 ) O(s − 1) O(s − 1) O(1)
= O(s − 1),
which contradicts Lemma 7.25. This concludes case I.
Case II: η is real character.
Lemma 7.26. Let m ∈ N+ . Then n|m η(n) ≥ 0. If m = l2 with
P
Since η is real, the only possible values for η(pj ) are 1, 0, and −1.
Corresponding to these cases, we observe that
rj + 1, if η(pj ) = 1,
1, if η(p ) = 0,
r j
η(1) + η(pj ) + · · · + η(pj j ) =
1, if η(pj ) = −1, and rj is even,
0, if η(pj ) = −1, and rj is odd.
P
Thus n|m η(n) is a product of non-negative factors. If m is a square,
P
all rj ’s are even, so that n|m η(n) is a product of numbers larger than
1.
−1
N η(n)
X N
X
sn n − (n + 1) + O(M −σ )
−σ −σ
=
n σ
n=M n=M
N
X −1
−σ
n − (n + 1)−σ + O(M −σ )
≤ φ(q)
n=M
φ(q) M −σ − N −σ + O(M −σ )
=
= O(M −σ ),
where the estimate |sn | ≤ φ(q) was demonstrated in the proof of The-
orem 7.24.
For N ∈ N+ , set
X η(n)
SN = √ .
m,n≥1; mn≤N
mn
The following two claims imply that L(1, η) 6= 0 and thus conclude the
proof of Theorem 7.21.
Claim 1: SN ≥ c log N , for some c > 0.
Proof: Write
√
N X N b Nc
X η(n) X
−1/2
X X 1
SN = √ = k η(n) ≥ ,
k=1 mn=k
mn k=1 l=1
l
n|k
92 7. CHARACTERS
The second term on the last line is√O(1) by Lemma 7.27, and the third
term is O(1) since there are only N terms in the sum. The first term
is a truncation of the series for L(1, η), so we get
∞
√ X η(n)
SII = 2 N [L(1, η) − √ ] + O(1),
√ n
n= N +1
which equals √
2 N L(1, η) + O(1)
by another application of Lemma 7.27.
We have thus proved Theorem 7.21.
7.4. Exercises
1. Let q be in N+ . Prove that gcd(n, q) = 1 if and only if there
exists m ∈ N+ such that mn ≡ 1, mod q.
2. Prove Proposition 7.22. (Hint: It is easy if G is cyclic. Then
show that G\1 × G2 = Ĝ1 × Ĝ2 ).
7.5. Notes
Theorem 7.15 is from [Hel69]. Our proof of Dirichlet’s theorem is
from [SS03]. This theorem was where Dirichlet series were first used
(and, in honor of this, were named after Dirichlet).
CHAPTER 8
Zero Sets
The fact that the zero sets for H p (D) are independent of p follows
from inner-outer factorization. An analogous factorization theorem
does not hold for the polydisk and the zero sets for H p (Dn ) depend on
p when n > 1.
Precise descriptions of the zero sets for the Bergman space or the
Dirichlet space are not known.
95
96 8. ZERO SETS
Then, clearly,
H p (Ω1/2 ) ⊂ H∞
p
(Ω1/2 ),
and
p f (s)
f ∈ H∞ (Ω1/2 ) =⇒ ∈ H p (Ω1/2 ), for p > 1.
s
A deeper result is the following, which is a variant of Hilbert’s
inequality. See [Mon94] or [HLS97] for a proof.
Theorem 8.6. H2 ,→ H∞
2
(Ω1/2 ).
We define Hp by to be the completion of the set of all finite Dirichlet
series with respect to the norm
Z T 1/p
1 p
kf kHp := lim |f (it)| dt .
T →∞ 2T −T
Also, if we define
X
A(θ) := σk − 1/2 ,
θ<tk <θ+1
Question 8.11. How can the zero sets of H2 , Hw2 , etc. be classified?
CHAPTER 9
Interpolating Sequences
for all i.
(2) The sequence {λi }∞ ∞ m
i=1 is an interpolating sequence for H (D ).
(3) The sequence {λi }∞ i=1 is weakly separated and the associated
Grammian with respect to Lebesgue measure σ is bounded.
Then (1) implies (2) and (2) implies (3). Moreover the converse of
both these implications is false.
To prove Seip’s theorem, we need to compare Gleason differences
in different algebras. For the remainder of the section, we shall adopt
the following notation:
zj − wj
dm (z, w) = ρH ∞ (Dm ) (z, w) = max
1≤j≤m 1 − w̄j zj
s − u
ρ(s, u) = ρH ∞ (Ω0 ) (s, u) =
s + ū
ρm (s, u) = dm ((2−s , ., p−s
m ), (2
−u
, ., p−u
m ))
where k̃zn is the reproducing kernel for the derivative. But this is
because differentiation maps the Hardy space (modulo constants) iso-
metrically onto a weighted Bergman space.
There is another proof of the fact that H 2 (D) does not admit any
sampling sequence, using the fact that multiplication by z is isometric.
110 9. INTERPOLATING SEQUENCES
The same idea works for H2 . The reader is invited to recast that proof
for the Hardy space.
Proposition 9.32. There are no sampling sequences on H2 .
Proof: The multiplication operator MN −s is an isometry on H2 .
On the other hand, the sequence fN := MN −s f tends to 0 uniformly
P N (sn )|2
on compact sets in Ω1/2 . Thus n |fkk sn k
2 → 0 as N → ∞ and thus
2 2
cannot be comparable to kfN k = kf k .
A similar argument shows that “generalized sampling sequences”
involving f 0 (sn ) do not exists.
Question 9.33. Is there a sensible interpretation of “{sn }n is a
generalized sampling sequence for H2 ”? If so, how are these character-
ized?
9.5. Exercises
Exercise 9.34. Prove Equation (9.4). (Hint: use an automorphism
of Dm to move one point to the origin).
Exercise 9.35. Prove that any sequence that tends sufficiently
quickly to ∂D is an interpolating sequence for H ∞ (D).
Exercise 9.36. Fill in the details of the proof of (9.16).
Exercise 9.37. Prove that if (hi ) is the minimal dual system of
(gi ), then the inverse of the Gram matrix G = hgj , gi i is the matrix
hhj , hi i.
9.6. Notes
For a much more comprehensive treatment of interpolating se-
quences, we recommend the excellent monograph [Sei04] by K. Seip.
For a concise treatment for H ∞ (D) and H 2 (D), including Theorem 9.20,
see [Nik85].
Axler’s theorem 9.2 was proved for multipliers of the Dirichlet space
[Axl92], but the argument readily adapts to the stated version. Car-
leson’s theorem is in [Car58]. Seip’s paper [Sei09] contains much
more information on interpolating sequences for H∞ than Theorem 9.10
alone.
Necessary and sufficient conditions for a sequence to be interpo-
lating for H ∞ (D2 ) are given in [AM01], but they do not completely
resolve the issue. For example, the following is still open:
Question 9.38. If λn is strongly separated in H ∞ (D2 ), is it an
interpolating sequence?
9.6. NOTES 111
Composition operators
Appendix
Proof: Let
r = max |wi |.
i=1,...,N
∞
Define g ∈ H (D) by
z
g(z) := f ( (w1 , . . . , wN )).
r
Then kgk∞ ≤ 1, and g(0) = 0. Apply Schwarz’s lemma to g to conclude
|g(r)| ≤ r.
≤ M ε,
where M = supn∈N kfn k. Since ε was arbitrary, we conclude that fn →
f weakly.
(2) =⇒ (3) : Let K ⊂ X be compact, then by continuity of κ,
the set K̃ := {kw ; w ∈ K} is also compact. Fix ε > 0 and find a
finite ε-net {kw1 , . . . , kwm } in K̃. Find N ∈ N such that for all n > N
hfn − f, kwj i < ε holds for j = 1, . . . , m. Then for any w ∈ K and
n > N:
|fn (w) − f (w)| = |hfn − f, kw i|
≤ |hfn − f, kwi i| + |hfn − f, kw − kwi i|
≤ ε + kfn − f k · kkw − kwi k
≤ ε + 2M ε
= (2M + 1)ε,
for a suitable i (such i exists since {kw1 , . . . , kwm } is an ε-net). Since
ε > 0 was arbitrary, we conclude that fn → f uniformly in K.
(3) =⇒ (4) : Obvious.
(4) =⇒ (1) : Follow immediately, since (4) means that (1) holds
with S = {kw }w∈X
we get (11.11).
Finally, if the norm on H is the L2 (µ)-norm, then the inequality
Z Z
2 2
|φf | dµ ≤ kφk∞ |f |2 dµ
X X
means kMφ k ≤ kφk∞ .
121
122 BIBLIOGRAPHY
E(ε, f ), 63 T∞ , 43
H∞ p
(Ω1/2 ), 97 Tr , 45
H∞ p
(Ω1/2 , 97 Hw2 , 67, 69
2
K, 77 Hw , 66, 69, 75
QK , 72 d(k), 8
B, 42 dj (k), 8
Xq , 87 gi , 105
D, 69 mte -Abschnitt, 62
E, 105 z r(n) , 42
H2 , interpolating sequences, 107
2 `2 (G), 87
Hw , interpolating sequences, 107
H∞ , 101
Hp , 97
σb , 38
Mult (H), 118
σu , 38
Q, 42
uniformly local H p space on Ω1/2 , 97
Θ(x), 12
ζ̃(s): alternating zeta function, 25 Abel’s Summation by parts formula,
β, 55 4
E, 44 abscissa of absolute convergence, 25,
NK , 72 27
ε-translation number, 63, 64 abscissa of bounded convergence, σb ,
RT
−−T
, 33 38
Ωρ , 3 abscissa of convergence, 25, 26
H2 , 57, 69 abscissa of uniform convergence, σu ,
.: less than or equal to a constant 38
times, 27 alternating zeta function, 25, 37
µ(n), 6
µα , 66 Beurling’s question, 56, 74
π(x), 5 Birkhoff-Khinchin Theorem, 77
Bohr condition, 23
P, 2
Pk , 2 Carleson measure, 101
ρ, 77 Carlson’s theorem, 68, 75, 78
ρA , 100 character, 62
σa : abscissa of absolute cyclic vector, 74
convergence, 25
σc : abscissa of convergence, 25 Dirichlet character modulo q, 87
125
126 INDEX
infinite polydisk, 62
interpolating sequence, for a Hilbert
function space, 105
interpolation constant, 99
Inverse Fourier Transform, 29
Inverse Mellin Transform, 30
quasi-character, 10, 62
Rademacher sequence, 44
relatively dense, 63
reproducing kernel, 117
Riemann hypothesis, 1
Riemann zeta function, 1
Riesz basis, 55